OpenAI preprint, September 2026 · §3. Probability bounds for the cells
Lemma3.2:Bounds and monotonicity for the inverse Mills derivative in reflected coordinates
Gaussian loss decreases with nonnegative thresholds.
Hover over a blue or green term in the text, or a node in the graph, to find it in the other. Select it to read its
declaration both in Lean and in words.
The Setting
Gaussian cell probability bounds involve the upper-tail probability at a real threshold and a loss weighted by its square. Let and denote the standard Gaussian density and distribution function on .
The unnormalized Gaussian density and the Gaussian normalizing constant are given by (1), which also fixes the normalization of :(1)The unnormalized Gaussian tail is the integral of over the open upper half-line, as in (2): (2)✦The Gaussian upper tail is the probability function in (3): (3)✦
The inverse Mills ratio is the density divided by the upper tail, with either normalization in (4): (4)✦ The inverse Mills derivative is the function given in (5), and its reflected form is the function obtained by setting :(5)The Gaussian loss is the function in (6): (6)✦ The inverse Gaussian distribution function is defined on probabilities in .
The Claim
Theorem 1. Let be nonnegative real thresholds. The reflected inverse Mills derivative satisfies (7) and (8): (7)✦(8)✦For every real threshold , its upper bound is (9): (9)✦For the ordered nonnegative thresholds , the loss satisfies (10): (10)✦
In the paper, the result reads: is nonincreasing on and satisfies for every real , and is increasing on .
What Is Checked
The assertions are stated over the real Gaussian density and upper-tail integral, with the inverse Mills ratio and its derivative defined in reflected coordinates, and the loss defined using the normalized upper-tail probability.